CN115292532A - Remote sensing image domain adaptive retrieval method based on pseudo label consistency learning - Google Patents
Remote sensing image domain adaptive retrieval method based on pseudo label consistency learning Download PDFInfo
- Publication number
- CN115292532A CN115292532A CN202210729817.7A CN202210729817A CN115292532A CN 115292532 A CN115292532 A CN 115292532A CN 202210729817 A CN202210729817 A CN 202210729817A CN 115292532 A CN115292532 A CN 115292532A
- Authority
- CN
- China
- Prior art keywords
- remote sensing
- samples
- target domain
- sensing image
- sample
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 47
- 230000003044 adaptive effect Effects 0.000 title claims abstract description 29
- 238000009826 distribution Methods 0.000 claims abstract description 47
- 239000013598 vector Substances 0.000 claims abstract description 46
- 230000006870 function Effects 0.000 claims abstract description 32
- 238000012549 training Methods 0.000 claims abstract description 22
- 238000013528 artificial neural network Methods 0.000 claims abstract description 10
- 238000005457 optimization Methods 0.000 claims abstract description 7
- 238000000605 extraction Methods 0.000 claims description 37
- 238000013527 convolutional neural network Methods 0.000 claims description 19
- 230000002708 enhancing effect Effects 0.000 claims description 13
- 239000011159 matrix material Substances 0.000 claims description 12
- 230000006978 adaptation Effects 0.000 claims description 11
- 239000000126 substance Substances 0.000 claims description 9
- 238000010276 construction Methods 0.000 claims description 6
- 238000013508 migration Methods 0.000 claims description 6
- 230000005012 migration Effects 0.000 claims description 6
- 238000012216 screening Methods 0.000 claims description 6
- 238000002372 labelling Methods 0.000 claims description 4
- 238000013507 mapping Methods 0.000 claims description 3
- 238000005259 measurement Methods 0.000 claims description 3
- 230000009466 transformation Effects 0.000 claims description 3
- 230000000694 effects Effects 0.000 abstract description 8
- 238000010586 diagram Methods 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 238000004140 cleaning Methods 0.000 description 1
- 230000002860 competitive effect Effects 0.000 description 1
- 238000012937 correction Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 239000002360 explosive Substances 0.000 description 1
- 230000002349 favourable effect Effects 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 238000005065 mining Methods 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/53—Querying
- G06F16/532—Query formulation, e.g. graphical querying
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/55—Clustering; Classification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/58—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
- G06F16/583—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
- G06V10/765—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects using rules for classification or partitioning the feature space
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Databases & Information Systems (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Multimedia (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Software Systems (AREA)
- Computing Systems (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Library & Information Science (AREA)
- Medical Informatics (AREA)
- Mathematical Physics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Molecular Biology (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention relates to a remote sensing image domain adaptive retrieval method based on pseudo label consistency learning, which comprises the following steps: a) Acquiring a remote sensing image; b) Constructing input data, a triple convolution neural network and a loss function; c) And performing optimization training on the triple convolution neural network by using input data and combining the loss function, extracting the remote sensing image characteristic vector of the remote sensing image in the target domain by using the trained triple convolution neural network to form a characteristic library vector, extracting the query characteristic vector queried by the user, and comparing the query characteristic vector with the remote sensing image characteristic vector in the characteristic library vector to obtain the remote sensing image characteristic vector in the set similarity rank. The remote sensing image domain adaptive retrieval method based on pseudo label consistency learning is less influenced by the distribution difference of the target domain and the source domain, and has a good retrieval effect.
Description
Technical Field
The invention relates to the technical field of optical remote sensing image retrieval, in particular to a remote sensing image domain adaptive retrieval method based on pseudo label consistency learning.
Background
In recent years, the remote sensing image is easier to obtain due to the development of earth observation technology, and mass remote sensing image data create favorable conditions for the application in the fields of earth surface coverage classification, disaster assessment, environment monitoring, city planning and the like.
However, in the growing remote sensing image data, how to efficiently find the interested target or scene becomes a difficult problem. Therefore, remote sensing image retrieval receives more and more attention as a key technology for mining effective information from large-scale remote sensing data.
At present, the retrieval model based on the deep neural network obtains the most competitive retrieval effect in remote sensing image retrieval. Since deep learning is a data-driven algorithm, these models are trained using a large amount of labeled data in order to achieve a better retrieval result. However, the explosive growth of remote sensing images poses a serious challenge to the data labeling work, which not only requires a large amount of manpower and material resources, but also is impractical to label all the images. Therefore, how to utilize the existing labeled remote sensing image and improve the retrieval precision of the model on the unlabeled data is a key problem to be solved. However, the retrieval effect of directly transferring the trained retrieval model to the unmarked data set is not ideal due to the difference of factors such as sensors, shooting angles, shooting weather and seasons among different data sets, and in the prior art, a classifier is generally learned by using labeled source domain data, and the classifier learned by the source domain is used for a target domain through feature alignment. However, subject to the difference in the distribution of the target domain and the source domain, the effect of the prior class information of the source domain on the target domain decision boundary is limited, which may result in the decision boundary learned by the source domain not being able to distinguish the target domain.
In view of the above, a remote sensing image domain adaptive retrieval method based on pseudo label consistency learning needs to be designed.
Disclosure of Invention
The technical problem to be solved by the invention is to provide a remote sensing image domain adaptive retrieval method based on pseudo label consistency learning, which is little influenced by the distribution difference of a target domain and a source domain and has good retrieval effect.
In order to solve the technical problem, the invention provides a remote sensing image domain adaptive retrieval method based on pseudo label consistency learning, which comprises the following steps:
a) Acquiring a remote sensing image;
b) Constructing input data, a triple convolution neural network and a loss function;
c) And performing optimization training on the triple convolutional neural network by using the input data and combining the loss function, extracting the remote sensing image feature vector of the remote sensing image in the target domain by using the trained triple convolutional neural network to form a feature library vector, extracting a query feature vector queried by a user, and comparing the query feature vector with the remote sensing image feature vector in the feature library vector to obtain the remote sensing image feature vector in a set similarity rank.
Further, the step of constructing the input data comprises:
b11 Constructed to contain n s Source domain data of individual samplesAnd comprises n t Target domain data of individual samplesWherein the content of the first and second substances,indicating that the source domain data has a labeled sample,representing the target domain data without annotated sample,representing annotated samples with said source domain dataA corresponding label, andc is the number of image categories;
b12 Unlabeled sample for the target domain dataObtaining target domain weak enhancement samples by using inversion and shift data enhancement transformationAnd the target domain data is not marked with samplesMethod for generating severely distorted target domain strongly enhanced samples by utilizing random enhancement method
B13 Has labeled samples on the source domain dataAnd obtaining source domain weakly enhanced samples by using the flipping and shifting data enhancement transformation.
Further, the triple convolution neural network comprises a feature extraction part and a classification part, wherein the feature extraction part comprises a plurality of feature extraction networks, the structures and the parameters of the feature extraction networks are the same, the classification part comprises a plurality of classifiers, and the structures and the parameters of the classifiers are the same.
Further, the feature extraction network consists of a convolutional neural network pre-trained by an ImageNet (image network) dataset.
Further, the classifier is a layer of fully connected network for predicting the likelihood that the input data belongs to different classes.
Further, the output dimension of the classifier is consistent with the number of classes of the input data.
Further, the constructing step of the loss function includes:
b31 Constructing a classification loss, performing supervised learning on the source domain data, and constructing the classification loss of the source domain data based on cross entropy loss:
wherein L is CE For the classification loss function, p (x) s ) Representing annotated samples of said source domain dataThe function p (|) represents the probability distribution predicted by the classifier, x s Labeling samples for the source domain dataThe collection of (a) and (b),representing annotated samples of said source domain dataIs the probability of a different class;
b32 Constructing migration loss L based on similarity of different feature distributions measured by maximum mean difference MMD :
Wherein i represents the ith labeled sample of the source domain datan s Representing annotated samples of said source domain dataTotal number of (f) i s Indicating the ith labeled sample of the source domain dataIs characterized by n t Represents the total number of samples of the target domain data, j represents the jth target domain strong enhancement sample, phi is a mapping function, and samples of the source domain data are projected to a high-dimensional Hilbert spaceIn the Hilbert spaceCalculating a sample mean of the source domain data and a sample mean of the target domain as a measure of domain difference;
b33 Constructing pseudo label consistency loss to obtain pseudo label classification loss L with consistency regular enhancement PCE :
Wherein B represents the number of samples selected by one training, mu is the proportion of samples meeting the selection requirement of a set threshold value in the samples selected by one training, H function represents the cross entropy loss of two probability distributions, and x w Represent the target domain weakly enhanced samplesThe collection of (a) and (b),after screening for pseudo tags x w Probabilities of being predicted as different classes;
b34 Constructing minimum class confusion loss, determining the weight of a sample by using the value of the class probability distribution entropy of the target domain data, calculating a class confusion matrix according to the weighted sample of the target domain data, and combining the minimum class confusion loss to maximize the inter-class difference of the target domain data; entropy of the probability distributionComprises the following steps:
wherein, the first and the second end of the pipe are connected with each other,represents the jth sample in the target domain data,representing the probability that the jth sample in the target domain data belongs to the c class, wherein the weight of the sample of the target domain data is defined as:
wherein, W j Representing a weight of a jth sample in said target domain data for scaling the weight, W j The corresponding diagonal matrix is W, and the confusion matrix M is defined based on the diagonal matrix W cc′ Comprises the following steps:
wherein the content of the first and second substances,representing the probability of all samples in the samples selected by one training belonging to the c-th class, and minimizing class confusion loss L MCC Is defined as follows:
b35 The loss function constructed is:
L=L CE +L MMD +αL PCE +βL MCC
wherein, alpha and beta are parameters for balancing the optimization target of the triple convolutional neural network.
Further, the step of training the triplet convolutional neural network by the input data comprises:
c11 Weakly enhancing the target domain samplesThe target domain strongly enhanced sampleAnd the source domain data has labeled samplesRespectively inputting the data into the corresponding feature extraction networks to obtain the target domain weakly enhanced sample featuresTarget domain strongly enhanced sample featuresWith source domain data having labeled sample features
C12 Strongly enhancing sample features with the target domainWith source domain data having labeled sample featuresPerforming distribution difference measurement to calculate the migration loss L MMD ;
C13 Weakly enhancing the target domain with sample featuresThe target domain strongly enhances sample featuresAnd the source domain has labeled sample featuresInputting the classifier to convert into a target domain weakly enhanced sample conditional probability distributionTarget domain strongly enhanced sample conditional probability distributionConditional probability distribution of labeled samples with source domain data
C14 Conditional probability distribution of labeled samples to the source domain dataBased on the classification loss function L CE Calculating the classification loss of the source domain data;
c15 Weakly enhancing sample conditional probability distribution to the target domainAnd (3) reserving the class label with the maximum class probability higher than the set probability distribution threshold value tau as a pseudo label:
wherein the content of the first and second substances,representing a strongly enhanced sample conditional probability distribution of the target domainThe category where the maximum probability of the set threshold value screening condition is met;
c16 Pseudo-label generated with the target domain weakly enhanced samplesAdopting the pseudo label classification loss L as supervision information of corresponding target domain strong enhancement samples PCE Calculating the consistency loss of the pseudo label;
c17 For the target domain weakly enhanced sample conditional probability distribution, utilizing the minimized class confusion loss L MCC Calculating class confusion loss;
c18 Computing total loss of training, and adjusting network parameters of the feature extraction network by using a gradient descent algorithm.
Further, the remote sensing image feature vector is obtained through the trained feature extraction network.
Further, the step of obtaining the remote sensing image of the target domain comprises:
c21 Extracting the query feature vector image based on the trained feature extraction network;
c22 Calculating Euclidean distance between the query feature vector image and each remote sensing image feature vector one by one;
c23 According to the sequence of the Euclidean distances from small to large, sorting the remote sensing image feature vectors, and taking the remote sensing image feature vectors ranked in a set order as high-similarity images.
According to the technical scheme, in the remote sensing image domain adaptive retrieval method based on pseudo tag consistency learning, input data are constructed firstly and comprise source domain data and target domain data, wherein the source domain data are labeled data, the target domain data are unlabeled, the constructed input data are correspondingly enhanced, then, the enhanced source domain data and the enhanced target domain data are input into a corresponding feature extraction network and a classifier, an output result is compared with the input data, pseudo tag consistency loss is established, accordingly, a loss function is obtained based on the pseudo tag consistency loss, network parameters of the feature extraction network can be adjusted based on the loss function, accordingly, the influence of distribution difference of the target domain and the source domain on the feature extraction network can be reduced, and the trained triple convolutional neural network has better retrieval accuracy and better retrieval effect when an unlabeled sample is retrieved.
Further advantages of the present invention, as well as the technical effects of preferred embodiments, are further described in the following detailed description.
Drawings
FIG. 1 is a flow chart of the remote sensing image domain adaptive retrieval method based on pseudo label consistency learning of the invention;
FIG. 2 is a schematic diagram illustrating the principle of the remote sensing image domain adaptive retrieval method based on pseudo label consistency learning according to the invention;
FIG. 3 is a schematic diagram of a training process of a triple convolution neural network in the remote sensing image domain adaptation retrieval method based on pseudo label consistency learning according to the invention;
FIG. 4 is a schematic diagram of a retrieval process in the remote sensing image domain adaptation retrieval method based on pseudo label consistency learning.
Detailed Description
The following describes in detail embodiments of the present invention with reference to the drawings. It should be understood that the detailed description and specific examples, while indicating the present invention, are given by way of illustration and explanation only, not limitation.
As shown in fig. 1 and fig. 2, as an embodiment of the remote sensing image domain adaptive retrieval method based on pseudo tag consistency learning provided by the present invention, the method includes the following steps:
a) Acquiring a remote sensing image;
b) Constructing input data, a triple convolution neural network and a loss function;
c) And performing optimization training on the triple convolutional neural network by using input data and combining with a loss function, extracting the remote sensing image feature vector of the remote sensing image in a target domain by using the trained triple convolutional neural network to form a feature library vector, extracting a query feature vector queried by a user, and comparing the query feature vector with the remote sensing image feature vector in the feature library vector to obtain the remote sensing image feature vector in the set similarity ranking.
Specifically, in an embodiment of the remote sensing image domain adaptive retrieval method based on pseudo tag consistency learning provided by the present invention, the input data construction step includes:
b11 Construction of a structure containing n s Source domain data of individual samplesAnd comprises n t Target domain data of individual samplesWherein the content of the first and second substances,indicating that the source domain data has a labeled sample,representing the target domain data without annotated sample,representing annotated samples with source domain dataA corresponding label, andc is the number of image categories;
b12 Unlabeled sample for target domain dataObtaining target domain weak enhancement samples by using inversion and shift data enhancement transformationAnd does not mark the target domain dataMethod for generating severely distorted target domain strongly enhanced samples by utilizing random enhancement method
B13 Has annotated samples to the source domain dataAnd enhancing the transformation by using the flip and shift data to obtain the source domain weak sample.
Further, in an embodiment of the remote sensing image domain adaptive retrieval method based on pseudo tag consistency learning provided by the present invention, as shown in fig. 3, the triple convolutional neural network includes a feature extraction part and a classification part, where the feature extraction part includes a plurality of feature extraction networks (for example, data samples extracted as needed include target domain weakly enhanced samplesTarget domain data label-free sampleWith source domain data labeled samplesThree, the feature extraction network is set to be three to practiceEach feature extraction network corresponds to different data samples), and the structure and parameters of each feature extraction network are the same; the classification section includes a plurality of classifiers (e.g., data samples classified as desired include target domain weakly enhanced samplesTarget domain data label-free sampleAnnotated sample with source domain dataThree classifiers are set to realize that each classifier corresponds to different data samples), and the structure and parameters of each classifier are the same, wherein the feature extraction network comprises a convolutional neural network ({ conv1, conv2_ x, conv3_ x, conv4_ x, conv _5x }) pre-trained by an ImageNet (image network) data set, and a bottleneck layer, and therefore, the structure of the feature extraction network is: { conv1, conv2_ x, conv3_ x, conv4_ x, conv _5x, bottoming layer }, whose output feature size is 256 dimensions; the classifier is a layer of fully-connected network and is used for predicting the possibility that input data belong to different categories, and the output dimension of the classifier is consistent with the number of the categories of the input data.
Further, in an embodiment of the remote sensing image domain adaptive retrieval method based on pseudo tag consistency learning provided by the present invention, the step of constructing the loss function includes:
b31 For the source domain data, because the source domain data has corresponding labeling information, supervised learning can be carried out on the source domain data so as to ensure that the source domain data has labeled samplesCan be accurately identified, and particularly, classification loss of source domain data can be constructed based on cross entropy loss:
wherein L is CE For the classification loss function, p (x) s ) Representing annotated samples of source domain dataTrue probability distribution of (2), x s Annotating source domain data with samplesThe collection of (a) and (b),representing annotated samples of source domain dataIs the probability of a different class;
b32 Constructing migration loss L based on similarity of different feature distributions measured by maximum mean difference MMD :
Wherein i represents the labeled sample of the ith source domain datan s Annotated samples representing source domain dataTotal number of (f) i s Indicating labeled sample of ith source domain dataIs characterized by n t Represents the total number of samples of the target domain data, j represents the jth target domain strong enhancement sample, phi is a mapping function, and projects the samples of the source domain data to a high-dimensional Hilbert spaceIn Hilbert spaceCalculating a sample mean value of the source domain data and a sample mean value of the target domain to be used as a measure of the domain difference;
b33 Constructing pseudo-label consistency loss for pseudo-label consistency learning that constrains target domain samples, weakly enhancing samples using target domainThe generated pseudo label is used as a corresponding target domain strong enhancement sampleThe cross entropy loss is calculated by the supervision information to obtain the consistency regular enhanced pseudo label classification loss L PCE :
Wherein B represents the number of samples selected by one training, mu is the proportion of samples meeting the selection requirement of a set threshold in the samples selected by one training, H function represents the cross entropy loss of two probability distributions, and x w Representing target domain weakly enhanced samplesThe collection of (a) and (b),after screening for pseudo labels x w Probabilities of being predicted as different classes;
b34 Constructing minimum class confusion loss, determining the weight of a sample of the target domain data by using the value of the class probability distribution entropy of the target domain data, calculating a class confusion matrix according to the weighted sample of the target domain data, and combining the minimum class confusion loss to maximize the inter-class difference of the target domain data; utensil for cleaning buttockEntropy of body-to-ground, probability distributionComprises the following steps:
wherein, the first and the second end of the pipe are connected with each other,represents the jth sample in the target domain data,representing the probability that the jth sample in the target domain data belongs to the c class, and the weight of the sample of the target domain data is defined as:
wherein, W j Representing the weight of the jth sample in the target domain data, for scaling the weight, W j The corresponding diagonal matrix is W, and a confusion-like matrix M is defined based on the diagonal matrix W cc′ Comprises the following steps:
wherein the content of the first and second substances,representing the probability that all samples in the samples selected by one training belong to the c-th class, and minimizing the class confusion loss L MCC Is defined as:
b35 Constructed loss function is:
L=L CE +L MMD +αL PCE +βL MCC
wherein, alpha and beta are parameters for balancing the optimization target of the triple convolutional neural network.
Further, in an embodiment of the remote sensing image domain adaptive retrieval method based on pseudo tag consistency learning provided by the present invention, the step of training the triplet convolutional neural network with input data includes:
c11 Weakly enhancing the target domain samplesTarget domain strongly enhanced samplesWith source domain data labeled samplesRespectively input into corresponding feature extraction networks to obtain the target domain weakly enhanced sample featuresTarget domain strongly enhanced sample featuresWith source domain data having annotated sample features
C12 Exploit the strong enhancement of sample features in the target domainWith source domain data having annotated sample featuresPerforming distribution difference measurement to calculate migration loss L MMD ;
C13 Weakly enhancing the target domain sample featuresStrongly enhanced sample features in target domainWith source domain data having annotated sample featuresInputting corresponding classifier to convert corresponding into target domain weakly enhanced sample conditional probability distributionTarget domain strongly enhanced sample conditional probability distributionConditional probability distribution of labeled samples with source domain data
C14 Has a conditional probability distribution of labeled samples over source domain dataBased on the classification loss function L CE Calculating the classification loss of the source domain data;
c15 Weakly enhancing sample conditional probability distribution to target domainAnd (3) reserving the class label with the maximum class probability higher than the set probability distribution threshold value tau as a pseudo label:
wherein, the first and the second end of the pipe are connected with each other,representing strongly enhanced sample conditional probability distributions for a target domainThe category where the maximum probability of the set threshold value screening condition is met;
c16 Pseudo-label generated with target domain weakly enhanced samplesAdopting pseudo label classification loss L as supervision information of corresponding target domain strong enhancement sample PCE Calculating the consistency loss of the pseudo label;
c17 For weakly enhanced sample conditional probability distribution in the target domain, with minimized class confusion loss L MCC Calculating class confusion loss;
c18 Computing total loss of training, and adjusting network parameters of the feature extraction network by using a gradient descent algorithm.
And after the training of the feature extraction network is completed, extracting the feature vector of the remote sensing image by the feature extraction network trained by the steps.
Further, in an embodiment of the remote sensing image domain adaptive retrieval method based on pseudo tag consistency learning provided by the present invention, as shown in fig. 4, the step of obtaining the remote sensing image of the target domain is as follows:
c21 Extracting a query feature vector image based on the trained feature extraction network;
c22 Calculating the Euclidean distance between the query feature vector image and the feature vector of each remote sensing image one by one;
c23 The remote sensing image feature vectors are sorted according to the sequence of Euclidean distances from small to large, and the remote sensing image feature vectors with ranks in a set order (such as the first K, the specific value of K can be manually specified, for example, the first 4) are taken as high-similarity images.
Further, in an embodiment of the remote sensing image Domain Adaptive retrieval method based on pseudo tag consistency learning provided by the present invention, the construction of the input data and the training of the triple convolutional neural Network are implemented based on a pytorreh library of Python language, and in addition, simulation experiments of retrieval systems such as ADDA (Adaptive discrete Domain Adaptation), AFN (Adaptive Feature Norm), BSP (Batch singular value constraint), CDAN (Conditional Adaptive Domain Adaptation) and DAN (Deep Adaptation Network) are also performed to compare with the remote sensing image Domain Adaptive retrieval method based on pseudo tag consistency learning of the present invention; the invention adopts an Average Normalized Modified Retrieval Rank (ANMRR), an average retrieval precision (mAP) and PK (K is the retrieval precision of the previous K images) to evaluate the result, wherein the higher the average retrieval precision (mAP) and the retrieval precision PK values of the previous K images are, the better the retrieval performance is, the smaller the ANMRR value is, the better the comparison result is shown in the following table 1:
TABLE 1
The results in the table 1 show that the remote sensing image domain adaptive retrieval method based on pseudo label consistency learning obtains the highest retrieval precision, and compared with a comparison method, the average retrieval precision mAP of the method is improved by 20.04-28.96%, and the average normalized correction retrieval rank ANMRR is also improved. In addition, the retrieval precision of P5-P100 of the method is superior to that of a comparison method, and in conclusion, the remote sensing image domain adaptive retrieval method based on pseudo label consistency learning can improve the retrieval capability of the query image of the target domain under the condition that the target domain is not labeled.
According to the technical scheme, in the remote sensing image domain adaptive retrieval method based on pseudo tag consistency learning, input data are constructed firstly and comprise source domain data and target domain data, wherein the source domain data are labeled data, the target domain data are unlabeled, the constructed input data are correspondingly enhanced, then, the enhanced source domain data and the enhanced target domain data are input into a corresponding feature extraction network and a classifier, an output result is compared with the input data, pseudo tag consistency loss is established, accordingly, a loss function is obtained based on the pseudo tag consistency loss, network parameters of the feature extraction network can be adjusted based on the loss function, accordingly, the influence of distribution difference of the target domain and the source domain on the feature extraction network can be reduced, and the trained triple convolutional neural network has better retrieval accuracy and better retrieval effect when an unlabeled sample is retrieved.
The preferred embodiments of the present invention have been described in detail above with reference to the accompanying drawings, but the present invention is not limited thereto. Within the scope of the technical idea of the invention, numerous simple modifications can be made to the technical solution of the invention, including any suitable combination of specific features, and in order to avoid unnecessary repetition, the invention will not be described in detail in relation to the various possible combinations. Such simple modifications and combinations should be considered within the scope of the present disclosure as well.
Claims (10)
1. The remote sensing image domain adaptive retrieval method based on pseudo label consistency learning is characterized by comprising the following steps of:
a) Acquiring a remote sensing image;
b) Constructing input data, a triple convolution neural network and a loss function;
c) And performing optimization training on the triple convolutional neural network by using the input data and combining the loss function, extracting the remote sensing image feature vector of the remote sensing image in the target domain by using the trained triple convolutional neural network to form a feature library vector, extracting a query feature vector queried by a user, and comparing the query feature vector with the remote sensing image feature vector in the feature library vector to obtain the remote sensing image feature vector in a set similarity rank.
2. The remote sensing image domain adaptation retrieval method based on pseudo-label consistency learning of claim 1, wherein the construction step of the input data comprises the following steps:
b11 Construction of a structure containing n s Source domain data of individual samplesAnd comprises n t Target domain data of individual samplesWherein the content of the first and second substances,indicating that the source domain data has a labeled sample,representing the target domain data without annotated sample,representing annotated samples with said source domain dataA corresponding label, andc is the number of image categories;
b12 Unlabeled sample for the target domain dataObtaining target domain weak enhancement samples by using inversion and shift data enhancement transformationAnd the target domain data is not marked with samplesMethod for generating severely distorted target domain strongly enhanced samples by utilizing random enhancement method
3. The remote sensing image domain adaptive retrieval method based on pseudo tag consistency learning of claim 2, wherein the triple convolution neural network comprises a feature extraction part and a classification part, wherein the feature extraction part comprises a plurality of feature extraction networks, the structures and the parameters of the feature extraction networks are the same, the classification part comprises a plurality of classifiers, and the structures and the parameters of the classifiers are the same.
4. The remote sensing image domain adaptation retrieval method based on pseudo-label consistency learning of claim 3, wherein the feature extraction network is composed of a convolutional neural network pre-trained by ImageNet data sets.
5. The remote sensing image domain adaptation retrieval method based on pseudo-label consistency learning of claim 3, wherein the classifier is a layer of fully connected network for predicting the possibility that the input data belongs to different categories.
6. The remote sensing image domain adaptation retrieval method based on pseudo-label consistency learning of claim 5, wherein the output dimension of the classifier is consistent with the number of categories of the input data.
7. The remote sensing image domain adaptation retrieval method based on pseudo-label consistency learning of claim 6, wherein the construction step of the loss function comprises the following steps:
b31 Constructing a classification loss, performing supervised learning on the source domain data, and constructing the classification loss of the source domain data based on cross entropy loss:
wherein L is CE For the classification loss function, p (x) s ) Representing annotated samples of said source domain dataThe function p (|) represents the probability distribution predicted by the classifier, x s Labeling samples for the source domain dataThe collection of (a) and (b),representing annotated samples of said source domain dataIs the probability of a different class;
b32 Constructing migration loss L based on similarity of different feature distributions measured by maximum mean difference MMD :
Wherein i represents the ith labeled sample of the source domain datan s Representing annotated samples of said source domain dataTotal number of (c), f i s Indicating the ith labeled sample of the source domain dataIs characterized by n t Represents the total number of samples of the target domain data, j represents the jth target domain strong enhancement sample, phi is a mapping function, and samples of the source domain data are projected to a high-dimensional Hilbert spaceIn the Hilbert spaceCalculating a sample mean of the source domain data and a sample mean of the target domain as a measure of domain difference;
b33 Constructing pseudo label consistency loss to obtain pseudo label classification loss L with consistency regular enhancement PCE :
Wherein B represents the number of samples selected by one training, mu is the proportion of samples meeting the selection requirement of a set threshold in the samples selected by one training, H function represents the cross entropy loss of two probability distributions, and x w Represent the target domain weakly enhanced samplesThe collection of (a) and (b),after screening for pseudo tags x w Predicated on different classes of generalitiesThe ratio;
b34 Constructing minimum class confusion loss, determining the weight of a sample by using the value of the class probability distribution entropy of the target domain data, calculating a class confusion matrix according to the weighted sample of the target domain data, and combining the minimum class confusion loss to maximize the inter-class difference of the target domain data; the probability distribution entropyComprises the following steps:
wherein the content of the first and second substances,represents the jth sample in the target domain data,representing the probability that the jth sample in the target domain data belongs to the c class, wherein the weight of the sample of the target domain data is defined as:
wherein, W j Representing a weight of a jth sample in said target domain data for scaling the weight, W j The corresponding diagonal matrix is W, and the confusion matrix M is defined based on the diagonal matrix W cc′ Comprises the following steps:
wherein the content of the first and second substances,representing the probability of all samples in the samples selected by one training belonging to the c-th class, and minimizing class confusion loss L MCC Is defined as:
b35 The loss function constructed is:
L=L CE +L MMD +αL PCE +βL MCC
wherein, alpha and beta are parameters for balancing the optimization target of the triple convolutional neural network.
8. The remote sensing image domain adaptation retrieval method based on pseudo-label consistency learning of claim 7, wherein the step of training the triplet convolutional neural network by the input data comprises:
c11 Weakly enhancing the target domain samplesThe target domain strongly enhanced sampleAnd said source domain data has labeled samplesRespectively inputting the data into the corresponding feature extraction networks to obtain the target domain weakly enhanced sample featuresTarget domain strongly enhanced sample featuresWith source domain data having annotated sample features
C12 Strongly enhancing sample features with the target domainWith source domain data having annotated sample featuresPerforming distribution difference measurement to calculate the migration loss L MMD ;
C13 Weakly enhancing the target domain sample featuresThe target domain strongly enhances sample featuresAnd the source domain data has labeled sample featuresInputting the classifier to convert into a target domain weakly enhanced sample conditional probability distributionTarget domain strongly enhanced sample conditional probability distributionConditional probability distribution of labeled samples with source domain data
C14 Conditional probability distribution of labeled samples to the source domain dataBased on the classification loss function L CE Calculating the classification loss of the source domain data;
c15 Weakly enhancing sample conditional probability distribution to the target domainAnd (3) reserving the class label with the maximum class probability higher than the set probability distribution threshold value tau as a pseudo label:
wherein the content of the first and second substances,representing a conditional probability distribution of strongly enhanced samples of the target domainThe category where the maximum probability of the set threshold value screening condition is met;
c16 Pseudo-label generated with the target domain weakly enhanced samplesAdopting the pseudo label classification loss L as supervision information of corresponding target domain strong enhancement samples PCE Calculating the consistency loss of the pseudo label;
c17 For the target domain weakly enhanced sample conditional probability distribution, utilizing the minimized class confusion loss L MCC Calculating class confusion loss;
c18 Computing total loss of training, and adjusting network parameters of the feature extraction network by using a gradient descent algorithm.
9. The remote sensing image domain adaptive retrieval method based on pseudo-label consistency learning of claim 8, wherein the remote sensing image feature vector is obtained through the trained feature extraction network.
10. The remote sensing image domain adaptive retrieval method based on pseudo label consistency learning of claim 9 is characterized in that the step of obtaining the remote sensing image of the target domain is as follows:
c21 Extracting the query feature vector image based on the trained feature extraction network;
c22 Calculating Euclidean distance between the query feature vector image and each remote sensing image feature vector one by one;
c23 According to the sequence of the Euclidean distances from small to large, sorting the remote sensing image feature vectors, and taking the remote sensing image feature vectors ranked in a set order as high-similarity images.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210729817.7A CN115292532B (en) | 2022-06-24 | 2022-06-24 | Remote sensing image domain adaptive retrieval method based on pseudo tag consistency learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210729817.7A CN115292532B (en) | 2022-06-24 | 2022-06-24 | Remote sensing image domain adaptive retrieval method based on pseudo tag consistency learning |
Publications (2)
Publication Number | Publication Date |
---|---|
CN115292532A true CN115292532A (en) | 2022-11-04 |
CN115292532B CN115292532B (en) | 2024-03-15 |
Family
ID=83821188
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202210729817.7A Active CN115292532B (en) | 2022-06-24 | 2022-06-24 | Remote sensing image domain adaptive retrieval method based on pseudo tag consistency learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN115292532B (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116188995A (en) * | 2023-04-13 | 2023-05-30 | 国家基础地理信息中心 | Remote sensing image feature extraction model training method, retrieval method and device |
CN117456312A (en) * | 2023-12-22 | 2024-01-26 | 华侨大学 | Simulation anti-fouling pseudo tag enhancement method for unsupervised image retrieval |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106980641A (en) * | 2017-02-09 | 2017-07-25 | 上海交通大学 | The quick picture retrieval system of unsupervised Hash and method based on convolutional neural networks |
CN112131967A (en) * | 2020-09-01 | 2020-12-25 | 河海大学 | Remote sensing scene classification method based on multi-classifier anti-transfer learning |
CN112699892A (en) * | 2021-01-08 | 2021-04-23 | 北京工业大学 | Unsupervised field self-adaptive semantic segmentation method |
CN113190699A (en) * | 2021-05-14 | 2021-07-30 | 华中科技大学 | Remote sensing image retrieval method and device based on category-level semantic hash |
US20210390686A1 (en) * | 2020-06-15 | 2021-12-16 | Dalian University Of Technology | Unsupervised content-preserved domain adaptation method for multiple ct lung texture recognition |
CN113889228A (en) * | 2021-09-22 | 2022-01-04 | 武汉理工大学 | Semantic enhanced Hash medical image retrieval method based on mixed attention |
CN114549909A (en) * | 2022-03-03 | 2022-05-27 | 重庆邮电大学 | Pseudo label remote sensing image scene classification method based on self-adaptive threshold |
-
2022
- 2022-06-24 CN CN202210729817.7A patent/CN115292532B/en active Active
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106980641A (en) * | 2017-02-09 | 2017-07-25 | 上海交通大学 | The quick picture retrieval system of unsupervised Hash and method based on convolutional neural networks |
US20210390686A1 (en) * | 2020-06-15 | 2021-12-16 | Dalian University Of Technology | Unsupervised content-preserved domain adaptation method for multiple ct lung texture recognition |
CN112131967A (en) * | 2020-09-01 | 2020-12-25 | 河海大学 | Remote sensing scene classification method based on multi-classifier anti-transfer learning |
CN112699892A (en) * | 2021-01-08 | 2021-04-23 | 北京工业大学 | Unsupervised field self-adaptive semantic segmentation method |
CN113190699A (en) * | 2021-05-14 | 2021-07-30 | 华中科技大学 | Remote sensing image retrieval method and device based on category-level semantic hash |
CN113889228A (en) * | 2021-09-22 | 2022-01-04 | 武汉理工大学 | Semantic enhanced Hash medical image retrieval method based on mixed attention |
CN114549909A (en) * | 2022-03-03 | 2022-05-27 | 重庆邮电大学 | Pseudo label remote sensing image scene classification method based on self-adaptive threshold |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116188995A (en) * | 2023-04-13 | 2023-05-30 | 国家基础地理信息中心 | Remote sensing image feature extraction model training method, retrieval method and device |
CN116188995B (en) * | 2023-04-13 | 2023-08-15 | 国家基础地理信息中心 | Remote sensing image feature extraction model training method, retrieval method and device |
CN117456312A (en) * | 2023-12-22 | 2024-01-26 | 华侨大学 | Simulation anti-fouling pseudo tag enhancement method for unsupervised image retrieval |
CN117456312B (en) * | 2023-12-22 | 2024-03-12 | 华侨大学 | Simulation anti-fouling pseudo tag enhancement method for unsupervised image retrieval |
Also Published As
Publication number | Publication date |
---|---|
CN115292532B (en) | 2024-03-15 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111191732B (en) | Target detection method based on full-automatic learning | |
CN110598029B (en) | Fine-grained image classification method based on attention transfer mechanism | |
CN110443143B (en) | Multi-branch convolutional neural network fused remote sensing image scene classification method | |
CN106909924B (en) | Remote sensing image rapid retrieval method based on depth significance | |
CN113378632B (en) | Pseudo-label optimization-based unsupervised domain adaptive pedestrian re-identification method | |
CN113190699B (en) | Remote sensing image retrieval method and device based on category-level semantic hash | |
Lin et al. | RSCM: Region selection and concurrency model for multi-class weather recognition | |
CN110909820A (en) | Image classification method and system based on self-supervision learning | |
CN112819023B (en) | Sample set acquisition method, device, computer equipment and storage medium | |
CN115292532A (en) | Remote sensing image domain adaptive retrieval method based on pseudo label consistency learning | |
CN112668579A (en) | Weak supervision semantic segmentation method based on self-adaptive affinity and class distribution | |
CN109948742A (en) | Handwritten form picture classification method based on quantum nerve network | |
CN113032613B (en) | Three-dimensional model retrieval method based on interactive attention convolution neural network | |
CN115471739A (en) | Cross-domain remote sensing scene classification and retrieval method based on self-supervision contrast learning | |
CN112434628A (en) | Small sample polarization SAR image classification method based on active learning and collaborative representation | |
CN113806580B (en) | Cross-modal hash retrieval method based on hierarchical semantic structure | |
CN112597324A (en) | Image hash index construction method, system and equipment based on correlation filtering | |
CN114863091A (en) | Target detection training method based on pseudo label | |
CN115457332A (en) | Image multi-label classification method based on graph convolution neural network and class activation mapping | |
CN114579794A (en) | Multi-scale fusion landmark image retrieval method and system based on feature consistency suggestion | |
CN114357221A (en) | Self-supervision active learning method based on image classification | |
CN114329031A (en) | Fine-grained bird image retrieval method based on graph neural network and deep hash | |
CN113065520A (en) | Multi-modal data-oriented remote sensing image classification method | |
WO2024082374A1 (en) | Few-shot radar target recognition method based on hierarchical meta transfer | |
CN111079840A (en) | Complete image semantic annotation method based on convolutional neural network and concept lattice |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |